NLP Tools Multiwords (The MEDLINE N-Gram Set)...disease hodgkin • Hodgkin Disease • HODGKINS...

NLP ToolsMultiwords

(The MEDLINE N-Gram Set)

By: Dr. Chris J. Lu

The Lexical Systems Group

NLM. LHNCBC. CGSB

June, 2015

• Lexical Systems Group: http://umlslex.nlm.nih.gov

• The SPECIALIST NLP Tools: http://specialist.nlm.nih.gov

http://lsg3.nlm.nih.gov/LexSysGroup/Home/index.html

http://www.nlm.nih.gov/

http://lhncbc.nlm.nih.gov/

http://umlslex.nlm.nih.gov/

http://specialist.nlm.nih.gov/

Table of Contents

Introduction• Natural Language Processing (NLP)

• The SPECIALIST NLP Tools

Multiwords• Introduction

• The MEDLINE N-Gram Set

• Exclusive Filters – Distilled MEDLINE N-Gram Set

• Future Work

Questions

http://umlslex.nlm.nih.gov/lexBuild


http://umlslex.nlm.nih.gov/lexicon


http://umlslex.nlm.nih.gov/lexCheck


http://lexsrv3.nlm.nih.gov/LexSysGroup/Projects/numbers/current/


http://umlslex.nlm.nih.gov/textTools


http://umlslex.nlm.nih.gov/gSpell


http://lexsrv2.nlm.nih.gov/LexSysGroup/Projects/scrt/current/


http://umlslex.nlm.nih.gov/lexAccess


Natural Language Processing

Map terms to concepts (meaning) Challenges: many to many mapping:

Terms Concepts

• cold • Cold Temperature|C0009264• Common Cold|C0009443• Cold Therapy|C0010412• Cold Sensation|C0234192• …

• cold• Cold Temperature• Cold Temperatures• Cold (Temperature)• Temperatures, Cold• Low temperature• low temperatures• …

• Cold Temperature|C0009264

















NLP – Concept Mapping

Normalization:• A term might have many different variations, such as

inflectional variants, spelling variants, synonyms, abbreviations (expansions), cases, ASCII conversion, etc.

• Normalize different forms of a concept to a same formQuery Expansion:

• Expand a term to its equal terms, such as subterm substitution of synonyms, derivational variants, spelling variants, abbreviations, etc.

• To increase recall

POS tagger:• Assign part of speech to a single word or multiword in a text.• To increase precision

















NLP - Norm

disease hodgkin

• Hodgkin Disease

• HODGKINS DISEASE

• Hodgkin's Disease

• Disease, Hodgkin's

• HODGKIN'S DISEASE

• Hodgkin's disease

• Hodgkins Disease

• Hodgkin's disease NOS

• Hodgkin's disease, NOS

• Disease, Hodgkins

• Diseases, Hodgkins

• Hodgkins Diseases

• Hodgkins disease

• hodgkin's disease

• Disease;Hodgkins

• Disease, Hodgkin

• …

•C0019829

•Hodgkin Disease

normalize

Indexed Database Normalized String

Index

Terms in Corpus

















NLP – Norm (Cont.)

normQuery Normed

termdisease hodgkin

Hodgkin’s Disease

Indexed Database

Normalized String

SQL

Results that matchesthe normalized query• C0019829• Hodgkin Disease

















NLP – Query Expansion

perforated ear drum

drum ear perforate

Norm

perforation ear drum

drum ear perforation

Norm

None C0206504Tympanic Membrane Perforation

Indexed Database Normalized String

















LVG - Lexical Variants Generation

To increase recall & precision

Query Expansion(Recall)

POS Tagging(Precision)

Inputs perforated ear drum saw

UMLS-CUI None • C1947903|verb|see• C0183089|noun|saw (device)

Process perforation ear drum noun

UMLS-CUI C0206504 • C0183089

Preferred Term Tympanic Membrane Perforation saw (device)

















NLP Tools by LSG

Lexical Systems Group: http://umlslex.nlm.nih.gov

The SPECIALIST NLP Tools: http://specialist.nlm.nih.gov



















The SPECIALIST NLP Tools

The SPECIALIST LEXICON Text Tools

Lexical ToolsLexBuild

LVG

NLP Applications























http://umlslex.nlm.nih.gov/lvg

http://umlslex.nlm.nih.gov/lvg











http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/TC/current/index.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/TC/current/index.html

Lexicon Coverage – by Word Count

Total word count for MEDLINE (2014): 2,725,710,505 Lexicon covers ~98% from MEDLINE

Types Word Count Percentage % Accu. %

LEXICON 2,542,758,048 93.2879% 93.2879%

NUMBER 7,797,019 0.2861% 93.5740%

DIGIT 126,635,190 4.6460% 98.2200%

MULTIWORD 18,549,715 0.6805% 98.9005%

NEW 29,970,533 1.0995% 100.0000%

Total 2,725,710,505

* Using Element Words to Generate (Multi)Words for the SPECIALIST Lexicon

Lu, Chris J.; Tormey, Destinee; McCreedy, Lynn; and Browne, Allen C.

AMIA 2014 Annual Symposium, Washington, DC, November 15-19, 2014, p. 1499

















http://lsg3.nlm.nih.gov/LexSysGroup/Docs/Papers/2014/14_AMIA_MW.pdf

Lexicon Coverage - by Unique Word

Total unique word for MEDLINE (2014): 3,264,205 Lexicon covers 11 ~ 12 % words in MEDLINE The rest of ~87.5% is the long tail (multiwords)

Types Word Count Percentage % Accu. %

LEXICON 291,271 8.9232% 8.9232%

NUMBER 61 0.0019% 8.9251%

DIGIT 75,406 2.3101% 11.2352%

MULTIWORD 42,045 1.2881% 12.5233%

NEW 28,55,422 87.4768% 100.0000%

Total 3,264,205

















Frequency Spectrum of MEDLINE 2014

















Words in Lexicon

Part of speech, inflection, lexical meaning

• saw|noun|singular|E0054443

• saw|verb|infinitive|E0054444

• saw|verb|past|E0055007

High frequency co-occur words?collocation is a sequence of words or terms that co-occur more often than would be expected by chance

• “non”, DC: 46,138, WC: 46,139• “study was to”, DC: 592,752, WC:593,718• “undergoing cardiac surgery”, DC: 2,589, WC: 3,135• “adverse cardiac”, DC: 4,405, WC:5,725

• “in the house”, DC: 1,170, WC: 1,298• in house|adj|positive|E0555310, DC: 1,681, WC: 2,129

















Single Words vs. Multiwords

Words include single words and multiwords Word boundary – space or tab Multiwords are words hat happen to be spelled with a space Single words vs. multiwords

• One word can be represented as a single word or multiword (clubfoot)

Single words Multiwords

saw club foot

ice-cream ice cream

clubfoot drop-foot gait

club-foot Horner’s syndrome

















Multiwords

A multiword is a word contains space(s)

Multiwords are used extensively in biomedical domain

Multiwords are an essential ingredient and play a key role for the

success of NLP task

Precise recognition of word boundaries and identify multiwords

benefit disambiguation and improves the accuracy in information

extraction

















Lexicon.2015

484,628 lexical records 1,070,220 words (categories and inflections) 896,213 forms (spelling only)

• Single words: 464,781 (51.86%); Multiwords: 431,432 (48.14%)

0

100000

200000

300000

400000

500000

600000

700000

800000

900000

1000000

02 03 04 05 06 07 08 09 10 11 12 13 14 15

Single Words Multiwords

















LexBuild Process

Built by linguists

LexBuild: a web-based computer-aided tool

Resources: a list of words (element words)• Add new lexical records if no exact/close match

• Update existing lexical records if related records are found by close match

• Multiwords that contain these words are reviewed through the Essie search engine*, Google Scholar, dictionaries, biomedical publications, domain-specific databases, nomenclature guidelines, and books, etc.

* N.C. Ide, R.F. Loane, D.D. Fushman, “Essie: A Concept-based Search

Engine for Structured Biomedical Text”, JAMIA, Vol. 14, Num. 3, May/June,

2007, p.253-263

















Element Word

Element words are lowercase single words without punctuation and are not stopwords

Single words/Multiwords Element words

• saw • saw

• ice-cream• ice cream

• ice• cream

• clubfoot • clubfoot

• club-foot• club foot

• club• foot

• drop-foot gait • drop• foot• gait

• Food and Drug Administration • food• drug• administration

















Stopwords

A high frequency words - preposition A grammar word –not too much meaning Examples:

o Lexical Tools (11): of, and, with, for, nos, to, in, by, on, the, (non mesh)

o Text Categorization (11,068): com, edu, htm, html, www, pdf, abandon, abandoned, etc.

















Element Word Approach - Issues

Use new element words (Frequency Rank: 1565 ~ 2549)

Time consuming

Frequency of element words (not multiwords)

New multiwords from old element words are missing

Essie search engine is not current

















Project Objective

A systematic way to add multiwords form MEDLINE to the SPECIALIST Lexicon• Covers high frequency multiwords

• High precision multiword candidate list

















N-Gram Model Approach

Get all N-Grams from MEDLINE documents• No MEDLINE N-Gram set available for public

Filter out N-Grams that are invalid words• Exclusive Filter: focus on not to drop recall, and then

increase precision Retrieve word candidates by patterns, rules, etc.

• Inclusive filter: focus on precision Expert validation

• Very expensive, minimize manual process

To bridge the gap between N-grams (statistical co-occurrence) and our term-based Lexicon.

















N-Gram

An n-gram is a contiguous sequence of n items from a given sequence of text or speech• An n-gram of size 1 is referred to as a "unigram“

• Size 2 is a "bigram" (or a "digram");

• Size 3 is a "trigram".

• Larger sizes are sometimes referred to by the value of n, e.g., "four-gram", "five-gram", and so on.

Example:• to be or not to be

N = 1 Unigram to, be, or, not, to, be

N = 2 Bigram to be, be or, or not, not to, to be

N = 3 Trigram to be or, be or not, or not to, not to be

















N-Gram Requirements

Range of N:• Lexicon.2014

Length: 50 (> 99.5508%) for Lexicon.2014 Others: frequency (WC and DC)

N WC Accumulated WC

1 457,335 (52.2615%) 457,335 (52.2615%)

2 281857 (32.2089%) 739,192 (84.4704%)

3 93011 (10.6287%) 832,203 (95.0991%)

4 29905 (3.4174%) 862,108 (98.5165%)

5 8358 (0.9551%) 870,466 (99.4716%)

6 2846 (0.3252%) 873,312 (99.7968%)

… … …

















MEDLINE N-Gram Set Generation

2014 Release:• Collect titles and abstracts from 22,356,869 MEDLINE documents

• Tokenize titles and abstracts to 126,612,705 sentences

• Parse sentences into n-grams

• Print out: DC|WC|N-Gram

TI & AB N-Gram Table Print OutTokenization

















MEDLINE N-Gram Set

PMID- 961031

OWN - NLM

STAT- MEDLINE

DA - 19761020

DCOM- 19761020

LR - 20041117

PUBM- Print

IS - 0042-2835 (Print)

VI - 10

IP - 1

DP - 1976 Jan-Feb

TI - Postoperative arrhythmias in open-heart surgery, A study on fifty cases.

PG - 30-7

AB - 50 consecutive patients undergone open heart surgery were analyzed

regarding postoperative arrhythmias in the first postoperative 3 days.

Disturbances of rhythm occurred in each case of our group, serious or not

serious (100%). Ventricular premature beats were the most frequent type of

arrhythmia in the first and second postoperative days (80%). Two cases

expired postoperatively. In one of them complete atrioventricular block

developed after double valvular replacements (mitral and tricuspid). The

other died of low cardiac output syndrome. Etiology of the arrhythmias

...


















MEDLINE N-Gram Set

PMID- 961031

OWN - NLM

STAT- MEDLINE

…

DP - 1976 Jan-Feb


PG - 30-7



Disturbances of rhythm occurred in each case of our group, …

…

JT - Vascular surgery

JID - 0103277


















MEDLINE N-Gram Set

PMID- 961031






S-ID Sentences

1 Postoperative arrhythmias in open-heart surgery, A study on fifty cases.

2 50 consecutive patients undergone open heart surgery were analyzedregarding postoperative arrhythmias in the first postoperative 3 days.

3 Disturbances of rhythm occurred in each case of our group, …

… … | …

















MEDLINE N-Gram Set

PMID- 961031






Apply sentence tokenizer on TI and AB• Check Ending

o Ends with “.”, “?”, “!”

o Not abbreviation, U. S. Army

o …

• Check Beginning

o Starts with Upper case, digit,

o Not “:”, “-”, “ “, “\t”

o ...

Unrecognized sentence pattern (~0.01%)

















MEDLINE Unigram


S-ID Sentences- PMID: 961031


2 50 consecutive patients undergone open heart surgery were analyzedregarding postoperativearrhythmias in the first postoperative 3 days.


4 …

Key Value (DC, WC)

postoperative 1 3

arrhythmias 1 2

in 1 3

open-heart 1 1

surgery 1 1

a 1 1

study 1 1

on 1 1

… … …

















MEDLINE N-Gram Set - Unigram

2014 Release:• For unigram, tokenized sentences into 2,610,209,406 words (tokens) and

saved in key-value table

Key (N-Gram) Value (WC, DC, PMID)

of 17,804,182 | 125,085,304

the 15,719,615 | 119,808,656

in 15,495,583 | 70,413,258

and 15,357,675 | 88,271,268

to 12,532,576 | 44,859,301

… … | …

















MEDLINE Bigram


S-ID Sentences


2 50 consecutive patients undergone open heart surgery were analyzedregarding postoperative arrhythmias in the first postoperative 3 days.


… … | …

Key DC, WC

postoperative arrhythmias 1 2

arrhythmias in 1 2

in open-heart 1 1

open-heart surgery 1 1

surgery, a 1 1

A study 1 1

study on 1 1

… … …

















MEDLINE N-Gram Set - Issues


N-grams N No. of N-Grams

unigrams 1 21,530,469

bigrams 2 205,868,398

trigrams 3 703,148,136

fourthgrams 4 1,295,096,308

fifthgrams 5 1,665,248,566

Takes too long because of big data (for N = 3)

Exceeds the limit for N >= 4

Max. keys in Java HashMap is 230 -1 (~109)

















MEDLINE N-Gram Set

Approach – Split, Group, Filter, Combine, and Sort*

Fourthgrams (automatic):• Split MEDLINE documents into 12 sections

and get the fourthgrams for each section

• Group fourthgrams from all 12 sections with

specified (10) alphabetic range, such as a-c,

c-e, e-f, etc.

• Apply WC (> 30) filter on all 10 groups

• Combine all 10 alphabetic ranges groups to

N-Gram set

• Sort

Split

Group

Filter

Combine

Sort

* AMIA 2015: Generating the MEDLINE N-Gam Set

Lu, Chris J.; Tormey, Destinee; McCreedy, Lynn; and Browne, Allen C.

















MEDLINE N-Gram Set - Specifications

N-grams 2014 2015

MEDLINE files 1-746 1-779

Max. length 50 50

Min. WC 30 30

MiN. DC 1 1

Total document 22,356,869 23,343,329

Total sentence 126,612,705 134,834,507

Total tokens 2,610,209,406 2,786,085,158

















MEDLINE N-Gram Set

N-grams 2014 2015

unigrams 804,382 843,206

bigrams 4,587,349 4,845,965

trigrams 6,287,536 6,702,194

fourthgrams 3,799,377 4,082,612

fifthgrams 1,545,175 1,674,715

N-Gram Set 17,023,819 18,148,692

Available to public

















Exclusive Filter - WC

2014 MEDLINE N-Gram Set

N-grams N No. of N-GramsNo. of n-grams

(WC >= 30)Pass Rate

unigrams 1 21,530,469 804,382 3.74%

bigrams 2 205,868,398 4,587,349 2.23%

trigrams 3 703,148,136 6,287,536 0.89%

fourthgrams 4 1,295,096,308 3,799,377 0.29%

fifthgrams 5 1,665,248,566 1,545,175 0.09%

nGram Set 1-5 3,890,891,877 17,023,819 0.44%

















Filter

MEDLINE

N-Gram Set

Enhanced

MEDLINE

N-Gram Set

Filter efficiency = trap terms / total terms

Filter passing rate = pass-through terms / total terms

Good filters have high efficiency and accuracy

Accuracy Test: apply filters on Lexicon (valid word set)

• Accuracy = TP + TN / TP + TN + FP + FN

= TP / TP + FN ….. TN & FP are 0

= trap / total terms

= pass rate

Trap (not retrieved) Pass (retrieved)

Valid (relevant) FN TP

Invalid (not relevant) TN FP

















A distilled N-Gram set by filtering out invalid words.

Applied high accuracy filter (V0 = V1 = … = Vn; I0 > I1 > … > In)

Higher precision with same recall rate (if filter has high accuracy rate)

N-Gram Precision n = Vn / (Vn + In)

= V0 / (V0 + In) ….. Vn is same as V0 (high accuracy)

> V0 / (V0 + I0) ….. I0 is bigger than In (high efficiency)

N-Gram Recall n = Vn / (Vn + FNn)

= Vn / (Vn + FN0) ….. FNn is a constant (0), same as FN0

= V0 / (V0 + FN0) ….. Vn is same as V0 (high accuracy)

MEDLINE

N-Gram Set

Enhanced

MEDLINE

N-Gram Set

N-Gram Filter-1 Filter-2 … Filter-N Distilled

Valid (TP) V0 V1 V2 … Vn Vn

Invalid (FP) I0 I1 I2 … In In

Serial Filters

















General

Filters (5)

Pattern

Filters (6)

Lead-End-Term

Filters (5)

Domain

Filters

ME

DLIN

E N

-Gra

m S

et

Dis

tille

d M

ED

LIN

E N

-Gra

m S

et

Distilled N-Gram Set

Enhanced M

ED

LIN

E N

-Gra

m S

et

















General Exclusive Filters

Filter Accuracy(875,890)

Pass RateN-Gram set

AccumulatedPass Rate

Trapped Examples

Pipe 100.0000% (0)

100.0000%(6)

100.0000% • 38|44|(|r| • 33|37|Ag|AgCl

Punctuation or space

100.0000%(0)

99.9977%(386)

99.9977% • 1259147|3690494|= • 604567|2377864|+/-

Digit 99.9999%(1)

99.3141%(116,772)

99.3118% • 1404799|2062240|2 • 239725|499064|95%

Number 99.9953%(41)

99.9760%(4,056)

99.2879% • 2463066|3359594|two • 18246|20674|first and

second

Digit and Stopword

99.9993%(6)

99.1595%(142,067)

98.4534% • 3155416|4125616|on the• 11180|12722|1, 2, and

















http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterPipe.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterPuncSpace.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterDigit.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterNumber.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterDigitStopword.html

Pattern Exclusive Filters


Pass RateN-Gram set


Trapped Examples

Parenthetic Acronym- (ACR)

100.0000% (0)

99.0232%(163,714)

97.4917% • 33117|33381|chain reaction (PCR)

• 30095|30315|polymerase chain reaction (PCR)

Indefinite article 99.9985% (13)

98.1703%(303,679)

95.7079% • 270384|292590|a case• 40271|40512|A series

UPPERCASE Colon 99.9999%(1)

99.4302%(92,841)

95.1625% • 2069343|2070116|RESULTS:• 18015|18016|AIM: The

Disallowed punctuation

99.9978%(19)

99.3020%(113,073)

94.4983% • 324405|719011|(n =• 86525|133350|(P < 0.05)

Measurement 99.9967%(29)

98.1947%(290,421)

92.7924% • 154905|181001|two groups• 12160|15197|10 mg/kg

Incomplete 99.9999%(1)

97.8470%(340,109)

90.7945% • 482021|1107869|(P • 25347|25992|years) with

















http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterParAcr.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterParAcr.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterIndefArtPat.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterColon.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterPuncDisallow.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterMeasurement.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterIncomplete.html

Lead-End-Terms Exclusive Filters


Pass RateN-Gram set


Trapped Examples

Absolute Invalid Lead-Term

99.9947% (46)

73.0945%(4,158,702)

66.3658% • 2780043|3451203|of a • 432921|434591|this study

was

Absolute Invalid End-Term

99.9997% (3)

78.8984%(2,384,059)

52.3615% • 1878109|3534031|patients with

• 1062545|1261445|between the

Lead-End-Term 99.9992%(7)

99.9741%(2,312)

52.3480% • 2578756|3106139|in a• 1733|1744|For one

Lead-Term no SpVar 99.9887%(99)

85.6678%(1,277,229)

44.8454% • 658430|708246|to determine• 533913|554628|In addition,

End-Term no SpVar 99.9975%(22)

83.1945%(1,283,001)

37.3089% • 1009451|1295670|number of • 726|734|(HPV) in

















http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterLeadTermAbs.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterEndTermAbs.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterLeadEndTermPat.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterLeadTermPat.html

http://lexlx1.nlm.nih.gov/LexSysGroup/Projects/lexicon/current/docs/designDoc/UDF/multiwords/exFilters/exFilterEndTermPat.html

Project Domain Exclusive Filters


Pass RateN-Gram set


Trapped Examples

Lexicon 100.0000%(0)

91.0478%(568,592)

33.9689% • 12532576|44859301|to• 44|44|systematic name

















Core-term

Strip initial and/or final punctuation from n-grams by

coreterm normalization• Strip initial chars if they are punctuation except for closed brackets

• Strip final chars if they are punctuation except for closed brackets

• Recursively strip close brackets of (), [], {}, <> at both ends

• trim

Input nGram Core-term

-in details in details

in details: in details

(in details:) in details

(in details:)) in details:)

(-(in details)%^) in details

{in (5) days}, in (5) days

((clean room(s))) clean room(s)

















Summary

Distributed MEDLINE N-Gram set (2014+) to public All exclusive filters have accuracy rate above 99.99%

(tested on Lexicon) Obtain the distilled MEDLINE N-Gram set at passing

rate of 37.30% • smaller data set• better precision• similar recall• used as baseline for further analysis

















Future Work

Inclusive Filters• Parenthetic Acronym Pattern

o computed tomography (CT)o magnetic resonance imaging (MRI)o polymerase chain reaction (PCR)o …

• EndWord Patternso Syndrome: migraine syndrome, contiguous gene syndrome, ...o Center: Heart Information Center, Veteran’s Affairs Medical Center,…o Disease: Fabry disease, Devic disease, …o …

• Spelling Variant Patterns (use distilled n-gram set)o SpVar normalizationo MES (Metaphone, Edit Distance, Sorted Distance)o ES (Edit Distance and Sorted Distance)

















Questions

Lexical Systems Group: http://umlslex.nlm.nih.gov

The SPECIALIST NLP Tools: http://specialist.nlm.nih.gov



















Date post:	05-Oct-2020
Category:	Documents
Upload:	others
View:	2 times
Download:	0 times

NLP Tools Multiwords (The MEDLINE N-Gram Set)...disease hodgkin • Hodgkin Disease • HODGKINS...

Documents